Search papers, labs, and topics across Lattice.
This paper analyzes LLM reasoning trajectories to show that failures often stem from a few critical early transitions, rather than being uniformly distributed. These transitions are marked by spikes in token-level entropy, suggesting points where the model's reasoning diverges. The authors introduce GUARD, an inference-time framework that leverages these uncertainty signals to probe and redirect critical transitions, improving reasoning reliability.
LLM reasoning often goes off the rails early, but targeted interventions at these critical junctures can dramatically improve performance.
Large Language Models (LLMs) achieve strong performance through extended inference-time deliberation, yet how their reasoning failures arise remains poorly understood. By analyzing model-generated reasoning trajectories, we find that errors are not uniformly distributed but often originate from a small number of early transition points, after which reasoning remains locally coherent but globally incorrect. These transitions coincide with localized spikes in token-level entropy, and alternative continuations from the same intermediate state can still lead to correct solutions. Based on these observations, we introduce GUARD, a targeted inference-time framework that probes and redirects critical transitions using uncertainty signals. Empirical evaluations across multiple benchmarks confirm that interventions guided by these failure dynamics lead to more reliable reasoning outcomes. Our findings highlight the importance of understanding when and how reasoning first deviates, complementing existing approaches that focus on scaling inference-time computation.
